OpenAI deploys Jalapeño ASICs with AMD EPYC Turin hosts, not Nvidia Vera
OpenAI is deploying its Jalapeño ASICs in rack-scale setups with AMD EPYC Turin hosts, each carrying 1.5TB of memory. Hardware VP Richard Ho called the Turin choice "pragmatic" and said Nvidia's Vera CPU is "a little bit behind on that maturity level."
- Each AMD EPYC Turin host in the Jalapeño rack carries 1.5TB of memory
- OpenAI called Turin a pragmatic, risk-reducing choice for the program
- Nvidia Vera was described as behind on platform maturity
- Turin is socketed while Vera is board-mounted
Read next
AI